drm/amdgpu:impl vgt_flush for VI(V5)

when shadowing enabled, tesselation app will trigger
vm fault because below three tesselation registers:

VGT_TF_RING_SIZE__CI__VI,
VGT_HS_OFFCHIP_PARAM__CI__VI,
VGT_TF_MEMORY_BASE__CI__VI,

need to be programed after vgt-flush.

Tesselation picture vm fault disappeared after vgt-flush
introduced.

v2:implement vgt-flush for CI & SI.
v3:move vgt flush inside of cntx_cntrl
v4:count vgt flush in frame_size
v5:squash in typo fix

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

3 files changed, 60 insertions, 25 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c
index 77b5918f606f..879a94bbfe12 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c
@@ -426,7 +426,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
                                 break;
-                        case 1: 
+                        case 1:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -446,7 +446,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
                                 break;
-                        case 3:  
+                        case 3:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -456,7 +456,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
                                 break;
-                        case 4:  
+                        case 4:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -466,7 +466,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 5:  
+                        case 5:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -476,7 +476,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 6:  
+                        case 6:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -486,7 +486,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 7:  
+                        case 7:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -496,7 +496,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
                                 break;
-                        case 8: 
+                        case 8:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -506,7 +506,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 9:  
+                        case 9:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -516,7 +516,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 10:  
+                        case 10:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -526,7 +526,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
                                 break;
-                        case 11:  
+                        case 11:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -536,7 +536,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 12:  
+                        case 12:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -546,7 +546,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 13:  
+                        case 13:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -556,7 +556,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 14:  
+                        case 14:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -566,7 +566,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 15:  
+                        case 15:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -576,7 +576,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 16:  
+                        case 16:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -586,7 +586,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 17:  
+                        case 17:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
@@ -596,7 +596,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 21:  
+                        case 21:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
@@ -606,7 +606,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 22:  
+                        case 22:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
@@ -616,7 +616,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
                                 break;
-                        case 23: 
+                        case 23:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
@@ -626,7 +626,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 24: 
+                        case 24:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
@@ -636,7 +636,7 @@ static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev)
                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
                                 break;
-                        case 25: 
+                        case 25:
                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
@@ -1463,6 +1463,13 @@ static void gfx_v6_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
         amdgpu_ring_write(ring, 0x1);
 }
 
+static void gfx_v6_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
+{
+        amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
+        amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
+                EVENT_INDEX(0));
+}
+
 /**
  * gfx_v6_0_ring_emit_hdp_invalidate - emit an hdp invalidate on the cp
  *
@@ -1917,7 +1924,7 @@ static int gfx_v6_0_cp_load_microcode(struct amdgpu_device *adev)
 
 static void gfx_v6_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
                                                bool enable)
-{       
+{
         u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
         u32 mask;
         int i;
@@ -2802,6 +2809,8 @@ static uint64_t gfx_v6_0_get_gpu_clock_counter(struct amdgpu_device *adev)
 
 static void gfx_v6_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
 {
+        if (flags & AMDGPU_HAVE_CTX_SWITCH)
+                gfx_v6_0_ring_emit_vgt_flush(ring);
         amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
         amdgpu_ring_write(ring, 0x80000000);
         amdgpu_ring_write(ring, 0);
@@ -3265,7 +3274,7 @@ static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_gfx = {
                 14 + 14 + 14 + /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */
                 7 + 4 + /* gfx_v6_0_ring_emit_pipeline_sync */
                 17 + 6 + /* gfx_v6_0_ring_emit_vm_flush */
-                3, /* gfx_v6_ring_emit_cntxcntl */
+                3 + 2, /* gfx_v6_ring_emit_cntxcntl including vgt flush */
         .emit_ib_size = 6, /* gfx_v6_0_ring_emit_ib */
         .emit_ib = gfx_v6_0_ring_emit_ib,
         .emit_fence = gfx_v6_0_ring_emit_fence,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index 06fddba54445..1a745cf93f47 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -2105,6 +2105,18 @@ static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
         amdgpu_ring_write(ring, 0x20); /* poll interval */
 }
 
+static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
+{
+        amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
+        amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
+                EVENT_INDEX(4));
+
+        amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
+        amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
+                EVENT_INDEX(0));
+}
+
+
 /**
  * gfx_v7_0_ring_emit_hdp_invalidate - emit an hdp invalidate on the cp
  *
@@ -2260,6 +2272,7 @@ static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
 
         dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
         if (flags & AMDGPU_HAVE_CTX_SWITCH) {
+                gfx_v7_0_ring_emit_vgt_flush(ring);
                 /* set load_global_config & load_global_uconfig */
                 dw2 |= 0x8001;
                 /* set load_cs_sh_regs */
@@ -5153,7 +5166,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
                 12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */
                 7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */
                 17 + 6 + /* gfx_v7_0_ring_emit_vm_flush */
-                3, /* gfx_v7_ring_emit_cntxcntl */
+                3 + 4, /* gfx_v7_ring_emit_cntxcntl including vgt flush*/
         .emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */
         .emit_ib = gfx_v7_0_ring_emit_ib_gfx,
         .emit_fence = gfx_v7_0_ring_emit_fence_gfx,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index ab84bff18727..a3684891c6e1 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -6186,6 +6186,18 @@ static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
         amdgpu_ring_write(ring, 0x20); /* poll interval */
 }
 
+static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
+{
+        amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
+        amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
+                EVENT_INDEX(4));
+
+        amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
+        amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
+                EVENT_INDEX(0));
+}
+
+
 static void gfx_v8_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
 {
         amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
@@ -6371,6 +6383,7 @@ static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
 
         dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
         if (flags & AMDGPU_HAVE_CTX_SWITCH) {
+                gfx_v8_0_ring_emit_vgt_flush(ring);
                 /* set load_global_config & load_global_uconfig */
                 dw2 |= 0x8001;
                 /* set load_cs_sh_regs */
@@ -6574,7 +6587,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
                 7 + /* gfx_v8_0_ring_emit_pipeline_sync */
                 128 + 19 + /* gfx_v8_0_ring_emit_vm_flush */
                 2 + /* gfx_v8_ring_emit_sb */
-                3, /* gfx_v8_ring_emit_cntxcntl */
+                3 + 4, /* gfx_v8_ring_emit_cntxcntl including vgt flush */
         .emit_ib_size = 4, /* gfx_v8_0_ring_emit_ib_gfx */
         .emit_ib = gfx_v8_0_ring_emit_ib_gfx,
         .emit_fence = gfx_v8_0_ring_emit_fence_gfx,